Interactions & Expression
Protein interactions, dimer interfaces, and cell-type expression.
Few proteins act alone. These panels place each isoform in its interaction and expression context — the dimer interfaces splicing can break, the wider network of partners a gene participates in, and where the gene is expressed across cell types and tissues.
Dimer Interface Disruption
Many proteins function as dimers — bound either to a second copy of themselves (homodimer) or to a distinct partner protein (heterodimer). A splice event that removes or reshapes the residues forming that binding interface can weaken or abolish the interaction, even when the rest of the protein looks intact.
Where the interface comes from
Interface residues are read from the AlphaFold Database multimer models of each complex. AFDB's summary endpoint lists every structure for a protein tagged with its oligomeric state (monomer, homodimer, or heterodimer); for each dimer we identify the residues that make contact across the interface in the canonical protein.
How residues are scored
Each canonical interface residue is projected onto the isoform through the same splice-aware alignment used elsewhere, then sorted into one of three states:
Preserved
Present in the isoform with essentially unchanged solvent accessibility — the contact is retained.
Perturbed
Still present, but its accessibility (SASA) has shifted — the contact may be conformationally compromised.
Absent
Mapped to a gap in the isoform — the residue was spliced out and the contact is physically gone.
The disruption score summarises this as (perturbed + absent) ÷ total interface residues, running from 0 (interface fully preserved) to 1 (fully disrupted). A separate flag marks interfaces where at least 50% of the residues are outright deleted, distinguishing wholesale loss of the interface from subtler conformational disruption.
Homodimer and heterodimer results are reported separately, and each heterodimer names its partner protein — so an isoform can be read as keeping some partnerships while losing others. On the isoform page, the interface residues can be projected straight into the Mol* structure viewer (“Visualise interface”) to see exactly which contacts survive the splice event.
STRING Interaction Network
Beyond direct dimers, each gene's highest-confidence partners are drawn from STRING (keyed by Ensembl protein ID). STRING's combined score is built from seven independent evidence channels:
Direct biochemical / biophysical assays.
Curated pathway & complex memberships.
Correlated expression across conditions.
Co-mention in the literature.
Co-occurrence across genomes.
Fused orthologs in other species.
Conserved genomic proximity (prokaryotes).
Note: STRING evidence is gene-level and does not distinguish isoform-specific interactions. A high combined score may reflect co-expression or literature co-mention rather than direct physical binding.
Expression — Cell Types & Tissues
Each gene carries a single-cell expression profile across 154 cell types from the Human Protein Atlas (HPA). Every entry pairs an expression value (nCPM) with a specificity category — for example cell-type enriched or low specificity — indicating whether the gene is broadly expressed or concentrated in a few cell types.
Note: HPA expression is measured at the gene level; isoform-specific abundance cannot be inferred directly from this profile.
Interactive tissue map
The top bar of every isoform page carries a human-body map that combines two layers:
- Long-read cohorts (coloured): tissues directly sampled by the SPLISOFORMS long-read cohorts (breast cancer, ccRCC kidney) are highlighted in bold colour with animated callouts.
- HPA baseline (grey): the HPA baseline organs are drawn statically underneath as spatial context for the expression profile.